Resource Management Algorithms Research Articles

Exploring the Impact of Resource Management Strategies on Simulated Edge Cloud Performance: An Experimental Study

Edge computing has emerged as a critical technology for meeting the needs of latency-sensitive applications and reducing network congestion. This goal is achieved mainly by distributing computational resources closer to end users and away from traditional data centers. Optimizing the utilization of limited edge cloud resources and improving the performance of edge computing systems requires efficient resource-management techniques. In this paper, we primarily discuss the use of simulation tools—EdgeSimPy in particular—to assess edge cloud resource management methods. We give a summary of the main difficulties in managing a limited pool of resources in edge cloud computing, and we go over how simulation programs like EdgeSimPy work and evaluate resource management algorithms. The scenarios we consider for this evaluation involve edge computing while taking into account variables like user location, resource availability, and network structure. We evaluate four resource management algorithms in a fixed, simulated edge computing environment to determine their performance regarding their CPU usage, memory usage, disk usage, power consumption, and latency performance metrics to determine which method performs better in a fixed scenario. This allows us to determine the optimal algorithm for tasks that prioritize minimal resource use, low latency, or a combination of the two. Furthermore, we outline areas of unfilled research needs and potential paths forward for improving the reliability and realism of edge cloud simulation tools.

Machine Learning Algorithms for Predictive Database Capacity Planning and Resource Management

In this work, the author examines the utilization of ready algorithmic tools of machine learning in the context of database medium and its capacity forecasting issues. They point to the growing need for ARCH-HERO scalable database systems that can easily adapt to changes in workload to maximize the usage of available resources. Studied simulation models only working with AI have illustrated the effectiveness of machine learning techniques for improving the functioning of a DBMS. Real-time case studies here expound on how ML models forecast database capacity needs, keep operational costs low, and enhance system efficiency. Other issues that concern DBMS and/or applying ML include challenges and prospects: the complexity of integration and/or embedding of ML into DBMS is presented along with potential solutions. This paper aims to share perspectives on the subsequent evolution of the databases utilized in ML and their ability to improve the resource management process.

Non-orthogonal multiple access-based MEC for energy-efficient task offloading in e-commerce systems

Mobile edge computing (MEC) reduces the latency for end users to access applications deployed at the edge by offloading tasks to the edge. With the popularity of e-commerce and the expansion of business scale, server load continues to increase, and energy efficiency issues gradually become more prominent. Computation offloading has received widespread attention as a technology that effectively reduces server load. However, how to improve energy efficiency while ensuring computing requirements is an important challenge facing computation offloading. To solve this problem, using non-orthogonal multiple access (NOMA) to increase the efficiency of multi-access wireless transmission, MEC supporting NOMA is investigated in the research. Computing resources will be divided into separate sub-computing that will be handled via e-commerce terminals or transferred to edge sides by reutilizing radio resources, we put forward a Group Switching Matching Algorithm Based on Resource Unit Allocation (GSM-RUA) algorithm that is multi-dimensional. To this end, we first formulate this task allocation problem as a long-term stochastic optimization problem, which we then convert to three short-term deterministic sub-programming problems using Lyapunov optimization, namely, radio resource allocation in a large timescale, computation resource allocating and splitting in a small-time frame. Of the 3 short-term deterministic sub-programming problems, the first sub-programming problem can be remodeled into a 1 to n matching problem, which can be solved using the block-shift-matching-based radio resource allocation method. The latter two sub-programming problems are then transformed into two continuous convex problems by relaxation and then solved easily. We then use simulations to prove that our GSM-RUA algorithm is superior to the state-of-the-art resource management algorithms in terms of energy consumption, efficiency and complexity for e-commerce scenarios.

Improved optimization algorithm for resource management in cloud applications with performance monitor of VM provisioning, placement and recycling

The machine learning technique has been used to increase cloud management’s intelligence. Effective resource provisioning also preserves the environment. Manual cloud management has some difficult problems, such as complexity in cloud systems and scale issues. Hence, this paper introduces a new task for managing the resources in the cloud using deep learning. The aim is to predict the overall workload and server status prediction to the cloud resource management. Initially, performance monitoring is performed to keep aware of the performance of the application and guarantee the cloud application’s performance. In the suggested work, the required data is collected for the resource utilization on multiple Virtual Machine (VM) metrics. The VM provisioning is performed next to rectify the issues of resource provisioning. After that, the workload and server status prediction is conducted, where the Weighted Recurrent Neural Network (W-RNN) is adopted. After attaining the predicted workload, the VM placement module is carried out. Here, the virtual resource’s quantity is attained. Moreover, the multi-objective functions like resource utilization; cost, energy, time, and Quality of Service (QoS) are derived in this phase with the help of the Improved Rain Optimization Algorithm (IROA). Subsequently, the VM recycling is performed in the suggested work. Here, a resource collector is given for the virtual resources recycling task. It scans the applications of the cloud in the data centre and processes the VM recycling for every application. While considering the statistical analysis of the IROA-W-RNN-based resource management system achieved a mean of 56.27% than JAYA-W-RNN, 21.09% than SCO-W-RNN, 60.2% than MFOA-W-RNN, and 16.74% than DA-W-RNN for configuration 4. Finally, the numerical analysis is conducted to validate the presented resource management task with the aid of various conventional tasks.

Self‐attention convolutional neural network optimized with Remora Optimization Algorithm for energy aware resource management in Non‐orthogonal Multiple Access networks

SummaryNon‐orthogonal multiple access (NOMA) is an efficient technology for future wireless communication systems. The resource management on NOMA networks is flattering for improving energy efficiency (EE) of systems. Therefore, self‐attention convolutional neural network (SACNN) optimized with Remora Optimization Algorithm (ROA) for energy aware resource management in NOMA network (SACNN‐ROA‐EE) is proposed in this paper for solving user allocation, channel allocation, and power allocation problems. The Matching‐Coalition approach is considered to face the user allocation issue, and the SACNN is considered to face the channel allocation and power allocation problems. SACNN does not adopt any optimization methods to determine the optimal parameters in channel allocation and power allocation. Therefore, the ROA is applied to optimize the SACNN weight parameters. The proposed technique is activated in MATLAB, and its efficacy is analyzed with under performances metrics, such as minimum user rate, sum rate, energy, and spectral performance. The proposed SACNN‐ROA‐EE attains 24.35%, 27.84%, and 38.58% better sum rate and 27.45%, 43.28%, 30.56% better EE compared with existing methods, such as EE utilizing communication deep neural network optimized with power allocation optimization (PAO‐CDNN‐EE), EE under deep transfer deterministic policy gradient along deep deterministic policy gradient approach (DPPGA‐DTDPPG‐EE), and EE utilizing deep reinforcement learning with baseline algorithms (BA‐DRL‐EE), respectively.

GLRM: Geometric Layout-Based Resource Management Method on Multiple Field Programmable Gate Array Systems

Multiple field programmable gate array (Multi-FPGA) systems are capable of forming larger and more powerful computing units through high-speed interconnections between chips and are beginning to be widely used by various computing service providers. However, the new computing architecture brings new challenges to the system’s task resource management. Existing resource management methods do not fully exploit resources in Multi-FPGA systems, and it is difficult to support fast resource request and release. In this regard, we propose a geometric layout-based resource management (GLRM) method for Multi-FPGA systems. First, a geometric layout-based task combination algorithm (TCA) was proposed to ensure that the final system can use the available FPGA resources more efficiently. Then, we optimised two resource management algorithms using TCA. Compared with state-of-the-art resource management methods, TCA increases resource flexibility by an average of 6% and resource utilisation by an average of 7%, and the two optimised resource management methods are effective in improving resource management performance.

Design of federated learning-based resource management algorithm in fog computing for zero-touch network

The concept of zero-touch networking involves creating networks that are fully autonomous and require minimal human intervention. This approach is increasingly relevant due to the rapid growth of current cloud architectures, which are beginning to reach their limits due to continuous expansion demands from users and within the network core itself. In response, Fog computing, acting as a smart, localized data center closer to network nodes, emerges as a practical solution to the challenges of expansion and upgrading in existing architectures. Fog computing complements cloud technology. However, the realization of zero-touch networks is still in its early stages, and numerous challenges hinder its implementation. One significant challenge is the NP-hard problem related to resource management. This paper introduces an optimal resource management algorithm based on Federated Learning. The effectiveness of this algorithm is evaluated using the iFogSim simulator within the existing cloud-fog architecture. The results demonstrate that the proposed architecture outperforms the current infrastructure in several key aspects of resource management, including system latency, number of resources processed, energy consumption, and bandwidth utilization.

Model order estimation method for user mobility classification in heterogeneous RAN systems

The performance of modern heterogeneous Radio Access Networks (RANs) can be improved by introducing new methods and algorithms for network resource management. In particular, user mobility information can be used to control RAN resources. To determine user mobility, information about the variation of the Signal to Interference + Noise Ratio (SINR) between base stations and users can be collected and analyzed. These data are multi-linear and can be represented as tensors to extract information from them. However, for the tensors decomposition, it is necessary to know the order of the multi-linear model or the rank of the tensors. Inspired by the LaRGE (LineAr Regression of Global Eigenvalues) method, we propose a novel method for the model order estimation of the multi-linear data based on the analysis of the Complimented Global Eigenvalues (CGE) profile. We have denoted our method as Enhanced LaRGE (EnLaRGE). We have compared the EnLaRGE method with the classical methods AIC, MDL and their multi-linear versions, as well as with the M-EFT and N-D EFT methods. EnLaRGE shows good performance for the tensors with small one of the tensors dimensions. EnLaRGE method is implemented for the automatic model order estimation of multi-linear data received from the Network Simulator 3 with RAN Intelligent Controller (RIC). The results show the high precision of the multi-linear model order estimation in comparison with state-of-the-art methods. The estimated model order can be used for user mobility classification and improving heterogeneous RANs performance. This research was funded by Russian Science Foundation, grant number 23-69-10084, 318, https://rscf.ru/project/23-69-10084/

Agent-based cloud simulation model for resource management

Driven by the successful service model and growing demand, cloud computing has evolved from a moderate-sized data center consisting of homogeneous resources to a heterogeneous hyper-scale computing ecosystem. This evolution has made the modern cloud environment increasingly complex. Large-scale empirical studies of essential concepts such as resource allocation, virtual machine migration, and operational cost reduction have typically been conducted using simulations. This paper presents an agent-based cloud simulation model for resource management. The focus is on how service placement strategies, service migration, and server consolidation affect the overall performance of homogeneous and heterogeneous clouds, in terms of energy consumption, resource utilization, and violation of service-level agreements. The main cloud elements are modeled as autonomous agents whose properties are encapsulated. The complex relationships between components are realized through asynchronous agent-to-agent interactions. Operating states and statistics are displayed in real time. In the evaluation, the efficiency of the simulator is studied empirically. The performance of various resource management algorithms is assessed using statistical methods, and the accuracy of server energy consumption models is examined. The results show that agent-based models can accurately reflect cloud status at a fine-grained level.

Research on the Influence of Electric Vehicle Integration in Island Microgrid, Vietnam

Vietnam's economy is developing strongly, and the demand for energy use will increase rapidly. The development of smart grids contributes significantly to the transition and sustainable development of energy from renewable energy sources to improve the quality of the national power supply and promote the sustainable use of electricity economically and efficiently. Thus, this is highly beneficial in reducing carbon emissions and other types of pollution. Besides, electrification in the transportation industry is developing rapidly, such as Electric Vehicles (EVs) and Metros in recent years. Integrating electric vehicles into the grid will enable two-way energy exchange, reactive power compensation and load balancing. However, the number of EVs participating in charging at a time will cause some conflicts, such as voltage and power loss at the nodes. Therefore, the balancing problem between load demand and generation source is a difficult task in planning operations. This paper presents a method to optimize island Microgrid (MG) operation with the participation of electric vehicles based on renewable energy sources. Optimization techniques in intelligent resource forecasting and management algorithms are built in MATLAB to achieve different requirements. The proposed Microgrid manages energy efficiency that adapts to the variability of Renewable Energy with improved efficiency.

A Taxonomy of Live Migration Management in Cloud Computing

Cloud Data Centers have become the key infrastructure for providing services. Instance migration across different computing nodes in edge and cloud computing is essential to guarantee the quality of service in dynamic environments. Many studies have been conducted on dynamic resource management involving migrating Virtual Machines to achieve various objectives, such as load balancing, consolidation, performance, energy-saving, and disaster recovery. Some have investigated to improve and predict the performance of single live migration. Recently, several research studies service migration in edge-centric computing paradigms. However, there is a lack of taxonomy and survey that focuses on the management of live migration in edge and cloud computing environments. In this article, we examine the characteristics of each field and propose a migration management-centric taxonomy to provide a holistic framework and guideline for researchers on the topic, including the performance and cost model, migration generations in resource management algorithms, migration planning and scheduling, and migration lifecycle management and orchestration. We also identify research gaps and opportunities to improve the performance of resource management with live migrations.

Network-Compute Co-Optimization for Service Chaining in Cloud-Edge-Radio 5G Networks

Flexible resource management can be facilitated by recent drastic advances in software-defined networking (SDN) and network function virtualization (NFV) technologies. Most of recent studies on resource management have mainly focused on the independent optimization of network function and computing function on top of a single-layer SDN architecture. In this paper, we focus on networking and processing resource orchestration to enhance the performance of network/computing functions in a hierarchical cloud-edge-radio 5G network architecture. Specifically, we propose a dynamic resource management algorithm, namely <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">RACER</i> , to process network/computing functions by opportunistically exploiting a tradeoff between the characteristics of the cloud and edge with respect to the amount of available resources and transmission delay. Herein, we decompose an original long-term problem including complex decisions in all network layers (that is, cloud, edge, and radio) into spatio-temporal subproblems without loss of optimality by leveraging Lyapunov optimization. Then, we derive the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">RACER</i> algorithm to find the decisions in each layer and each time slot. This online-fashioned algorithm is pragmatic because it not only exploits information obtained from instantaneous time slots, but also operates in a distributed manner with small message exchanges between the different layers. Moreover, we demonstrate the excellent performance improvement of the proposed dynamic and holistic <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">RACER</i> algorithm compared to static and independent policies in the 5G network environment.

Optimal Resource Management for Hierarchical Federated Learning Over HetNets With Wireless Energy Transfer

Remote monitoring systems analyze the environment dynamics in different smart industrial applications, such as occupational health and safety, and environmental monitoring. Specifically, in industrial Internet of Things (IoT) systems, the huge number of devices and the expected performance put pressure on resources, such as computational, network, and device energy. Distributed training of Machine and Deep Learning (ML/DL) models for intelligent industrial IoT applications is very challenging for resource limited devices over heterogeneous wireless networks (HetNets). Hierarchical Federated Learning (HFL) performs training at multiple layers offloading the tasks to nearby Multi-Access Edge Computing (MEC) units. In this paper, we propose a novel energy-efficient HFL framework enabled by Wireless Energy Transfer (WET) and designed for heterogeneous networks with massive Multiple-Input Multiple-Output (MIMO) wireless backhaul. Our energy-efficiency approach is formulated as a Mixed-Integer Non-Linear Programming (MINLP) problem, where we optimize the HFL device association and manage the wireless transmitted energy. However due to its high complexity, we design a Heuristic Resource Management Algorithm, namely H2RMA, that respects energy, channel quality, and accuracy constraints, while presenting a low computational complexity. We also improve the energy consumption of the network using an efficient device scheduling scheme. Finally, we investigate device mobility and its impact on the HFL performance. Our extensive experiments confirm the high performance of the proposed resource management approach in HFL over HetNets, in terms of training loss and grid energy costs.

High Throughput VMs Placement With Constrained Communication Overhead and Provable Guarantees

Placement of VMs in the cloud is one of the most fundamental problems in systems research. Traditionally, placement algorithms assume that the schedulers have complete information about the currently available resources at each host. However, this assumption is in many cases unrealistic, as gathering fresh status information from each of the thousands of hosts in a large data center incurs excessive communication overhead, which results in long queueing delays. Efforts to resolve this problem by employing several parallel schedulers typically exhibit collisions when several schedulers are simultaneously trying to place VMs on the same host. Our work analyzes the performance of various placement algorithms and provides empirical evidence that using multiple randomized schedulers obtains high throughput, while significantly decreasing both the communication overhead, and the number of collisions between schedulers. We, therefore, introduce Adaptive Partial State Random (APSR) – an efficient parallel random resource management algorithm that samples only from a small number of hosts and dynamically adjusts the degree of parallelism to provide provable guarantees on the probability of collisions between distinct schedulers. We formally analyze APSR, evaluate it on real workloads, and integrate it into the popular OpenStack cloud management platform. Our evaluation shows that APSR matches the throughput provided by other parallel schedulers, while achieving up to 13x lower decline ratio and a reduction of over 85% in communication overheads.

A joint cluster-based RRM and Low-latency framework using the full-duplex mechanism for NR-V2X networks

The radio resource management (RRM) problem in new radio vehicle-to-everything (NR-V2X) communication systems is a combinatorial optimization problem, and it is hard to achieve an optimum result in polynomial time. To reduce the complexity, linear algorithms or meta-heuristics can solve the problem. In this study, we proposed a joint cluster-based resource management and low-latency framework using a full-duplex mechanism (JCRRM-FD) for NR-V2X networks, which is used for vehicle-to-vehicle (V2V)/ device-to-device (D2D) and vehicle-to-infrastructure (V2I) transmission in NR-V2X networks. The delay, resource management, and system throughput analysis of the full-duplex mechanism in this framework was performed to deal with resource utilization and latency constraints. Moreover, joint cluster-based radio resource management and ant colony optimization (ACO) algorithms were proposed to manage and optimize resources efficiently to achieve user separation. The swarm intelligence algorithm is a standard meta-heuristic algorithm employed to deal with the optimization concern by exploiting the V2X communication network’s unlimited speed (maximizing the data transfer capacity and overall network performance) while considering the quality-of-service (QoS) requirements. A comprehensive experiment analysis was enacted to evaluate the efficiency of the developed JCRRM-FD framework with baseline approaches. Based on the simulation results, the proposed JCRRM-FD framework enhances the fairness index, average delay, packet drop rate, best cost value, CDF, and throughput compared to the benchmark approaches.

Two-stage clustering for improve indoor positioning accuracy

High accurate positioning, as a key factor for the most resource management algorithms, has attracted a lot of attention due to its crucial role in some new applications such as smart factories, IoT, and autonomous cars. Indoor positioning, as another major category of positioning in addition to outdoor positioning, is a very complicated process and achieving high accuracy is a tough process. This paper proposes a two-stage clustering-based approach (TSCA) for indoor positioning to deal with highly complex data sets which was collected from a large-scale indoor radio system. In the first stage, a new clustering algorithm, called group matching method is developed, which divides the dataset into several groups (sub-datasets) according to the different reference signal received power (RSRP) values of the real-world dataset. In the second stage, KNN and its variants, are used to match and evaluate the location of each device in one of sub-datasets instead of the entire dataset, which can increase the accuracy of the positioning. This method can perfectly solve the problem of uneven distribution of reference point data in the process of data acquisition, which is a popular challenge for most real scene data acquisition. The proposed method is compared with several state-of-the-art ML methods such as Support Vector Regression (SVR), and clustering methods such as K-means. The results indicate a high positioning accuracy improvement of more than 55% compared to a modified KNN method use our own RSRP fingerprint dataset, and a 2D accuracy improvement of 13.36% and a 3D accuracy improvement of 10.3% compared to a traditional KNN method use a Wi-Fi Received Signal Strength fingerprint.

Performance Evaluation of Hybrid Meta-Heuristics-Based Task Scheduling Algorithm for Energy Efficiency in Fog Computing

Task scheduling in fog computing is one of the areas where researchers are having challenges as the demand grows for the use of internet of things (IoT) to access cloud computing resources. Many resource scheduling and optimization algorithms were used by many researchers in fog computing; some used single techniques while others used combined schemes to achieve dynamic scheduling in fog computing, many optimization techniques were assessed based on deterministic and meta-heuristic to find out solution to task scheduling problem in fog computing but could not achieve excellent results as required. This article proposes hybrid meta-heuristic optimization algorithm (HMOA) for energy efficient task scheduling in fog computing, the study combined modified particle swarm optimization (MPSO) meta-heuristic and deterministic spanning tree (SPT) to achieve task scheduling with the intention of eliminating the drawbacks of the two algorithms when used separately, the MPSO was used to schedule user task requests among fog devices, while hybrid MPSO-SPT was used to perform resource allocation and resource management in the fog computing environment. The study implemented the proposed algorithm using iFogSim; the performance of the algorithm was evaluated, assessed, and compared with other state-of-the-art task scheduling and resource management algorithms, the proposed method performs better in terms of energy consumption, resource utilization and response time, and the study proposed future research on evaluating the execution time using the hybrid algorithm.

Performance evaluation of opportunistic schedulers based on fairness and throughput in new-generation mobile networks

This paper discusses radio resource management in the telecommunications industry that utilizes opportunistic scheduling approaches and methods. It proposes a new scheduler called IFMR (Improved Fairness of MaxRate) to extend the opportunistic approach with the proposition of a new scheduling solution that enables to significantly maximize the system throughput. Our goal is to ensure better results that maximize system throughput, ensure perfect fairness at three levels and maximize the number of satisfied users who have achieved their desired throughput. Similarly, the performance of the proposed algorithm is evaluated and compared with reference algorithms: round robin, MaxSNR and proportional fair (PF). The focus of this study is to evaluate and analyze the quality of service guaranteed to each network subscriber, considering the heterogeneous data traffic based on simulations. The algorithms presented in this paper are programmed and simulated using the MATLAB simulation software. The study aims to effectively analyze the impact of radio resource management algorithms and schedulers, including the round robin algorithm, MaxSNR and PF, among others, with the main objective of evaluating the performance of the schedulers used by telecommunication and cellphone network operators. The scheduling phase, which precedes resource allocation and assignment, is a critical aspect of the study. The impact of these schedulers on throughput and fairness will also be evaluated and analyzed based on specific criteria.

Neural network inspired differential evolution based task scheduling for cloud infrastructure

In recent years, cloud computing has become an essential technology for businesses and individuals alike. Task scheduling is a critical aspect of cloud computing that affects the performance and efficiency of cloud infrastructure. During this pandemic where most of the healthcare services like COVID-19 sampling, vaccination process, patient management and other services are dependent on cloud infrastructure. These services come with huge clients and server load in a small instance of time. These task loads can only be managed at cloud infrastructure where an efficient resource management algorithm plays an important role. The optimal utilization of cloud infrastructure and optimization algorithms plays a vital role. The cloud resources rely on the allocation policy of the tasks on cloud resources. Simple static, dynamic, and meta-heuristic techniques provide a solution but not the optimal solution. In such a scenario machine learning and evolutionary algorithms are only the solution. In this work, a hybrid model based on meta-heuristic technique and neural network is proposed. The presented neural network inspired differential evolution hybrid technique provides an optimal assignment of the tasks on cloud infrastructure. The performance of the DE-ANN hybrid approach is performed using performance metrics, average start time(ms), average finish time(ms), average execution time(ms), total completion time(ms), simulation time(ms), and average resource utilization respectively. The proposed DE-ANN approach is validated against BB-BC, and Genetic approaches. It outperforms the existing meta-heuristic techniques i.e. Genetic approach, and Big-Bang Big-Crunch. The performance is evaluated using two configuration scenarios using 5 virtual machines and 10 virtual machines with varying tasks from 1000 to 4500. Experimental results show that the DE-ANN technique significantly improves task scheduling performance compared to other traditional techniques. The technique achieves an average improvement of 19.15% in total completion time(ms), 32.23% in average finish time(ms), 51.95% in average execution time(ms), and 33.24% in average resource utilization respectively. The DE-ANN technique is also effective in handling dynamic and uncertain environments, making it suitable for real-world cloud infrastructures.

A GPU-accelerated adaptive particle refinement for multi-phase flow and fluid-structure coupling SPH

In this paper, a graphical processing unit (GPU) implementation of adaptive particle refinement (APR) in the Smoothed Particle Hydrodynamics (SPH) framework is used to simulate the multi-phase flow and fluid-structure coupling problems. The multi-phase flow and fluid-structure coupling SPH models based on the Riemann solver are presented and the local particle refinement technique based on the axis-aligned square refinement pattern is adopted to split particles. A particle shifting technique constructing virtual fine particles (VFP-PST) is proposed to regularize the particle distribution for numrical accuracy and stability. To facilitate the device memory throughput efficiently, a dynamic resource management algorithm is used to implement GPU-accelerated adaptive particle refinement. Precision analysis shows the effect of the smooth length ratio of fine and coarse particles on numerical accuracy. Different tests have demonstrated that the proposed GPU-accelerated IAPR is accurate and stable and provides a promising approach to simulate the multi-phase flow and fluid-structure problems with low computational cost and comparable precision when compared with uniform particles.